Tracking the UK "Second Wave" Death Toll

Adam Larson (aka Caustic Logic)

September 27, 2020

updates Dec. 8

As the United Kingdom sees a rise in Covid19 infections, PM Johnson and others have suggested a new round of stay-at-home "lockdown" order may become necessary. Many are frustrated and confused. For example, independent journalist Anna Brees, a lockdown skeptic, shows the UK deaths curve (still looking totally flat at the end, and asks "lock me down for this?" 

As I see it, the first round of lockdown was "for this" - to make the people make that the death curve, instead of the one they were on the way to. As this shows and everyone knows, daily deaths had been doing nothing but rising sharply before they leveled off and then fell rather suddenly, starting in mid-April. That's about 2 weeks after restrictions were imposed and new infections witnessed a similar, if milder, decline. Most logically that's less contact, making less infections, and thus less fatalities. If that had been delayed, deaths would have kept rising to even higher levels it would take longer to arrest and then climb down from. 

Of course, it's a "second wave" and the deaths from that they worry about, and these haven't happened yet, aside from perhaps the low-rumbling start of them. I had just been looking gain at UK deaths again and noticed they've been rising - they had been down 4 or 5 most days, sometimes zero. But now they've gone back up to 10, 15, 26 a day and rising, with revisions pushing tallies into the 30s, and the latest full day currently says 45 deaths just in England and Wales on September 23 (these are Public Health / NHS stats - ONS still tallies a higher number for England and Wales, I noticed - 51,917 there, to official UK total 42,060 including Scotland and Northern Ireland). 

That rise doesn't yet show up clearly on a scale distorted by highs around 1,000 deaths per day, but it's a bad sign, and just the start of what will wind up hurting. Having a look at recent infections, it's clear that second wave is real and has already been happening. Official stats when I checked a few days ago - note the last day is tiny, provisional - all these days at the end can change, mostly up and probably just by a bit. With that, it's been hitting up to 5,000 confirmed cases per day. (in fact one day now revised to 5,197, and the site currently says for daily cases not yet plotted: 6,042. They will need to expand their graph.

https://coronavirus.data.gov.uk/cases

Maybe all those freewheeling protests are having an impact? Maybe some herd-immunity vigilantes are even out there deliberately spreading it to save the UK from tyranny? Possible. But mainly it's just the communicability of this virus plus a lot more communing than was happening; regular commerce and social activity. But then see the photo below from Trafalgar Square yesterday, September 26 (Time). To the extent this spirit prevails, their activity is not being done carefully except where it's required - and it largely has been, which is what's got these people so frustrated they prove as loud and clear as possible that they need to be babysat, even as they demand not to be. Babies, after all, don't understand.

Already confirmed infections are near past April levels. That's partly due to wider testing, of course, so it's not really that big in comparison - yet. And I do not foresee near-April levels of deaths coming on (above 800/day for a month straight back then), nor even the same rate scaled down to current real infections. "Protecting the vulnerable" can never be done totally, but it has been done fairly well, and that's been the main thing, IMO, that lets deaths remain so low people can easily pretend the danger has gone away for good. But again, it can't be done totally, and with this scale of infections that we know of, more yet that we don't, and especially if higher levels are yet to come -it will be harder for those of us in serious danger to keep avoiding this extremely contagious pathogen.

If second waves happen for real and not just in British leaders' agendas - maybe France would see one?  2nd wave - cases now far above April levels, confirming up to 16,000 cases in a day, and deaths are back up to 40-50 a day since Sept. 15, already with three spike-catchup days of 80, 154, and 150 each (in a slightly smaller populace than the UK). https://www.bing.com/covid/local/france?vert=graph

Now the UK death toll remains almost flat so far, but of course it takes time to die from the virus, so trying to call the second wave death toll is clearly premature. Almost instantly on seeing cases near 5,000/day, I predict the daily deaths will rise to 50, then 100, then hitting triple-digit daily tolls regularly, just from what we see here. Depending how high it gets, they may again see 300 dead in a day or more. Depending how long it takes to rein that back in, it could add a hell of a lot more deaths along the way. 

Projecting 2 Weeks' Deaths

Premature or not, I tried for a data-based estimate, and was able to show my initial guess  (" triple-digit daily tolls regularly, just from what we see here") was high. I tried to line up dates between cases, hospitalizations, and deaths in various ways - something I hadn't tried yet and didn't know how to do. For reference, I used this drugs.com page with a timeline of general progression that lines up with the bits I've heard: first symptoms is day 1 - fever, cough, aches, etc. Around days 7-9 dyspnoea and sepsis set in, in bad cases. Day 12 or so, improvement begins or ARDS forms, patient is admitted to ICU.  "Day 15: Acute kidney and cardiac injury becomes evident" and then "Day 18.5: The median time it takes from the first symptoms of COVID-19 to death is 18.5 days." (which would make it day 19.5, I think, starting from 1)

First I rushed off with a comparison of cases to deaths 18 days later (giving a crude cfr of 1.28% for what it's worth). But confirmation doesn't happen at infection. I suppose just when varies ... but the above gives 6 days on average in ICU, in turn usually 12 days after infection.  ... I tried a lineup of the curves for cases, hosp, deaths, on the idea surges would tend to show up similarly at the right times, besides scattering to surrounding days and mixing with other days' scatter. But with all the local wiggles and uncertain variables, it was hard to say. Generally, having the three set about  a week apart makes for a basic fit. Somehow I decided to use: hospital 6 days after confirmation, death 9 days after that (just 15 days total span). A ways in I realized even this wasn't the best, but I don't have the combination of skills and time to get it right, and it doesn't matter much - the differences are only so big, and some will cancel each other out.

Over the most recent 23 reasonably complete days, I count: 

372 deaths up to 9/23 (from 9/1), out of 

37,979 cases up to 9/8 (corresponding days), and 

3,207 hospitalized up to 9/14 (corresponding days).

I calculate a 0.98% crude CFR (case-fatality ratio) - nearly 1% of known cases have died so far. That's rather low and more like an Infection-Fatality ratio (IFR), historically, but confirmation is probably much higher now than it has been, bringing IFR and CFR a bit closer to synonymous.)

8.44% hospitalization rate - some 92% are considered ok to quarantine at home, but 8% require intensive help

11.6% crude "hfr" (hospitalization-fatality rate) between the 23 paired days (deaths 6 days later) - nearly 12% who have to go to the hospital never leave.

Using this cfr and hfr compared to daily cases and hospitalizations from Sept. 25, I project a total of around 519 deaths set for the next 14 days, as broken down below. 

The "cfr" I get comparing deaths 15 days from cases seems fairly steady down the line, but my hospital fatality rates are increasing over time. Around Sept. 5/6 hospitalizations is where it starts rising faster, from about 9% (as set up) to as high as 17.7 and 19.5%. An average of last 12 days gives a 13% hfr, which seemed to better fit the actual rise in deaths. so I used that. In fact the latest death tolls might show I'm still estimating low. 

Hospitalizations themselves seem to increase in the same time, from 130/day for a week up to Sept. 7, then quickly to 164, 192, 192, 205, 198, 208, 231, 247, 248.  Why these rise? wrong day line-up causes an increased distortion, or was causing it, and the higher rate is more accurate? Or hospitalizations and the fatality rate of them are actually rising? It could be that as young and careless people people spread the virus wider, it's now be spreading more to the less eager citizens; they avoid it for so many days and then don't, wind up in the hospital at a higher rate, and then die there at a higher rate than we were seeing 2 weeks ago. 

With this set-up, the longest projection is 15 days out from the latest reported cases tally (again, subject to change). I used 0.98% for each day's cases, and 13% for each corresponding day's hospitalizations (where available - six days later runs out 6 days before the end), to estimated deaths 15 days out. I just averaged the cfr and hfr predictions, which tended to be quite close. The 13% hfr does pull it a bit higher than the cfr alone, adding about 4-5 deaths at the end. So I mimicked that, adding 4-5 from there. We'll see if the current hfr makes it lower or higher than I estimate. This suggests what my earlier try did for a peak of deaths so far at 55 in a day - and I emphasize - so far.

What I'll be Watching

I will come back to check my prediction and bring updates on the confirmed death toll, the reactions of the government and the public, and maybe a look at the unconfirmed death toll. Another post deals with overall UK "excess deaths" (besides sorting out the varied and changing official tallies) - speculative analysis of the true death toll based on when and how people die at above-average rates. many have noted some 30,000 disputed non-covid "excess deaths" above the average, disputed if caused by the virus, or lockdown, or what. It tends to happen right when Covid19 is killing a lot, and when we can see how, it's in mostly cardiovascular ways where the virus might well play a hand. I'm having another go at this, expanded in a few ways, including to track the recent rises in covid and overall deaths, now above average when they had been well below for about 2 months. And I'll bring some more specific clues that bear out my thesis those excess deaths were probably more coronavirus ones never confirmed - and likely even some of the deaths BELOW the average line are as well. I've found so far the true U.K. death toll is probably over 70,000, possibly  as high as 80,000. I'll see about refining that point at that post,  after another week might settle some trend confusion I'll have to explain - maybe after I understand it.  

October 19: I haven't followed all developments. But cases are now at around 18-20,000 per day, hospitalizations are 900-1,000/day, and deaths are higher than I predicted, hitting around 90 most days and so far maxing out on October 11 at 103 (England and Wales - 117 for all U.K.). As before, all days open to revision, mainly up, and especially in the last few days.

December 8 update: I did this to help explain some points, including a conservative prediction of 320+ deaths per day by mid-November. (note: 2,550 is no kind of upper end - I compromised to avoid re-sizing the image to be even taller.

It turns out that was quite conservative and again yielded low predictions. Cases had risen to 20,000 per day on average as London went into lockdown (tier 3 restrictions) October 17, only slowing a climb to 25,000 in early November (with peaks past 31k on Nov. 2 and Nov. 9), and then declining to about 15k by early December. Hospitalizations peaked around 1,800/day all through mid-November, decreasing only to 1,400 by early December. when lockdown was eased in London Dec. 2 (tier 3 to tier 2). The peak of deaths would be 2-4 weeks after the highest numbers of cases, so has probably already passed, but the data may be incomplete. So far the highest tallies are 433 on Nov. 9 and 452 on Nov. 18 (again, England and Wales - all UK: 481 on the 9th, 484 on the 18th). 

I may have finally guessed too high here, estimating 580 daily UK deaths peaking around Nov. 22/23, as if that angled line really matters. I suppose the peak will be right around then, but only about 500 when it settles not much higher than it is by now. 

Comparing covid to all-cause excess deaths (E & W) from the end of lockdown 1 to lockdown 2 and the span between. That adds a bit to the question of what caused all those deaths under lockdown, which I addressed here.